Magnetic random access memories (MRAMs) are expected to be nonvolatile memories which provide a high speed operation and an infinite number of rewritings, and vigorous developments thereof have been carried out. In an MRAM, a magnetoresistive element is integrated within a memory cell, and a data is stored as the orientation of the magnetization of the ferromagnetic layer of the magnetoresistive element. Although several approaches have been proposed as a method of switching the magnetization of the ferromagnetic layer, all of them are common with regard to the use of a current. In putting into the practical use, the reduction of the write current is very important, and there is a requirement lot the reduction down to 0.5 mA or less, preferably, to 0.2 mA or less, according to 2006 Symposium on VLSI circuits, Digest of Technical Papers, p. 136.
The most typical, one of data writing methods into an MRAM is to dispose an interconnection through which a write current is flown near the magnetoresistive element, and to switch the orientation of the magnetization of the ferromagnetic layer of the magnetoresistive element with a current magnetic field, which is generated by flowing the write current. This method is preferable in attaining the high speed MRAM, because the data write in one nanosecond or less can be achieved in principle. For example, Japanese Patent Publication No. P2005-150303A discloses a structure in which the magnetization of the end portion of a magnetization fixed layer is oriented in the film thickness direction, for an MRAM in which the data is written by the current magnetic field.
However, a magnetic'field necessary for switching the magnetization of magnetic material which has sufficient thermal stability and resistance against external magnetic field disturbance is typically several ten oersteds, and a large write current around several miliamperes is required to generate such a large magnetic field. When the write current is large, the chip area is inevitably increased and the power consumption necessary for writing is also increased, which causes poor competitiveness compared with other random access memories. In addition, when the size reduction of memory cells causes a further increase in the write current; this is undesirable from the viewpoint of scaling.
In recent years, the following two approaches have been proposed to solving such problems. The first approach is to use a spin injection magnetization reversal. In an MRAM using the spin injection magnetization reversal, a magnetoresistive element of a memory cell is provided with a film stack including a first ferromagnetic layer (often referred to as a magnetization free layer), a second ferromagnetic layer having a fixed magnetization (often, referred to as a magnetization fixed layer), and a tunnel barrier layer disposed between these ferromagnetic layers. In data writing into such an MRAM, the magnetization of the magnetization free layer is reversed by using an interaction caused between localized electrons in the magnetization free layer and spin-polarized conduction electrons when a current is sent between the magnetization free layer and the magnetization fixed layer. The occurrence of the spin injection magnetization reversal depends on the current density (not on the absolute value of the current), and thus, when the spin injection magnetization reversal is used in the data writing, the write current is decreased as the size of the memory cell is reduced. That is, the spin injection magnetization reversal can be said to be superior in the scaling property. However, a write current is required to flow through the tunnel barrier layer, which has a thin film thickness, in data writing. This causes a problem of rewriting durability and reliability. Also, a current path is commonly used in writing and reading, and this may cause an erroneous writing in reading. As thus described, there are several obstacles in attaining the practical use of the spin injection magnetization reversal, although the spin injection magnetization reversal is superior in the scaling property.
The second approach is to use the current-driven domain wall motion. The magnetization reversal using the current-driven domain wall motion allows solving the above-described problems caused by the spin injection magnetization reversal. An MRAM which uses the current-driven domain wall motion is disclosed in, for example, Japanese Patent Publications No. P2005-191032A, P2006-73930A, and P2006-270069A. In the most typical configuration of an MRAM which uses the current-driven domain wall motion, a ferromagnetic layer (often, referred to as a magnetic recording layer) for holding the data is provided with: a magnetization reversible portion having a reversible magnetization; and two magnetization fixed portions having fixed magnetizations and connected to the respective ends of the magnetization reversible portion. The data is stored as the magnetization of the magnetization reversible portion. The magnetizations of the two magnetization fixed portions are fixed to be approximately anti-parallel to each other. When the magnetizations are thus arranged, the domain wall is introduced into the magnetic recording layer. When a current is sent in the direction that passes through the domain wall, the domain wall is moved in the direction of conduction electrons, as reported in Physical Review Letters, vol. 92, number 7, p. 077205, (2004), and thus data can be written by sending a current through the magnetic recording layer. Also, the occurrence of the current-driven domain wall motion depends on the current density, and thus it can be concluded that the current-driven domain wall motion provides a good scaling property similarly to the spin injection magnetization reversal. In addition, the above-described problems of the spin injection magnetization reversal can be solved in a memory cell of an MRAM that uses the current-driven domain wall motion, since the write current does not flow though the insulating layer, and the write and read current paths are separately provided.
However, an MRAM that uses the current-driven domain wall motion suffers from a problem that the absolute value of the write current is relatively large. According to a large number of reports announced with regard to the observations of the current induction domain wall motion, the current density of about 1×10−8 [A/cm2] is necessary for the domain wall motion. In this case, the write current is 1 mA, when the width of the ferromagnetic film in which the domain wall motion occurs is 100 nm and the film thickness is 10 nm, for example. A further reduction in the write current may be achieved by decreasing the width and thickness of the ferromagnetic film; however, a fact is reported in which the current density necessary for the writing is further increased when the film thickness is reduced (for example, refer to Japanese Journal of Applied Physics, vol. 45, No. 5A, pp. 3850-3853, (2006)). Also, the reduction in the width of the ferromagnetic film down to 100 nm or less involves the severe difficulty from the viewpoint of the processing technique.
One promising approach for decreasing the current density is to use a film of vertical magnetic anisotropic material that has a magnetic anisotropy in the film thickness direction, as the magnetic recording layer (the layer in which the domain wall motion occurs). In the magnetoresistive element that uses the film made of the vertical magnetic anisotropic material, the threshold current density in the order of 106 [A/cm2] is observed [for example, refer to Applied Physics Letters, vol. 90, p. 072508 (2007)].
The use of the film made of vertical magnetic anisotropic material in the magnetic recording layer, however, makes it difficult to increase the magnetoresistance ratio (MR ratio), which corresponds to the SN ratio of the reading signal of the MRAM, resulting in a problem that the compatibility with the reading property is difficult. Recently, very large magnetoresistance ratios are reported mainly with respect to magnetic tunnel junctions having a structure of CoFeB/MgO/CoFeB; however, CoFeB is a material having a magnetic anisotropy in the in-plane direction. In addition, magnetic tunnel junctions have been developed with various materials; however, most of them are materials, having a magnetic anisotropy in the in-plane direction. As for the vertical magnetic anisotropic material, there are very few achievements in which a magnetic tunnel junction is obtained with a high magnetoresistance ratio and high reliability.